v1.10.7

Signal Arcade v1.10.7 improves learning correctness, market-burst handling and reliability during long-running paper-trading sessions.

What’s improved

• More timely learning evidence: checkpoints approaching their collection deadline receive priority within their evidence lane, while preserving fairness and provider budgets.
• Less wasted refresh work: invalid route identities no longer repeatedly consume lookup slots. Changed identities retry, and temporary provider failures remain eligible.
• Stronger Entry proof: coverage follows the current Baseline and feature contract. Durable Policy identities preserve training/proof separation after older evidence is pruned. Linear and XGBoost candidates are compared only across matching validation populations.
• Correct Coach progression: winning Coach contenders earn fresh activation proof and remain monitored under their active identity. Waiting Coach proposals are protected from unrelated native candidate updates.
• More efficient burst processing: removes duplicate held-position valuations and database writes, reduces repeated learning calculations, and processes queued urgent evidence in small transactions without waiting to collect a batch.
• Safer cancellation and recovery: submitted work completes within its ownership boundary, and training rechecks market pressure before preparing a fit.
• Reliable long-season accounting: removes the 100,000-fill accounting limit and keeps Results and Seasons consistent across fills and season rollover.
• Better diagnostics: adds timing evidence for worker CPU, execution delays and contention. Diagnostics retain their separate 512 MiB allowance.
• Updated documentation: refreshed README, validation notes and 18 live screenshots in a dedicated v1.10.7 folder.

Learning and safety

Baseline boundaries, Champion permissions, Linear/XGBoost training recipes, independent proof and the 70% coverage requirement remain intact. Missing outcomes remain visible rather than being excluded to improve qualification.

These changes improve efficiency and evidence handling; they do not guarantee an Entry Champion or profitable trading.

Upgrading

Use Settings → Maintenance & updates → Prepare for upgrade, wait for Ready, and back up your data before updating.

This release introduces database schema 16. Existing bankroll, positions, seasons, learning records and Champion history are retained; no new season is required.

Migrations are forward-only. Rolling back to an older schema requires restoring its matching pre-upgrade data backup and image together.

Docker image: nicxx2/signal-arcade:1.10.7

Signal Arcade remains a paper-only simulator.

v1.10.6

Signal Arcade v1.10.6 improves automatic Champion support, explains how AI Coach research can contribute, and brings clearer charts, season history and more responsive market processing.

🏆 Independent Champion support

• Entry, Manipulation, Sizing and Exit can qualify independently. Entry no longer needs to qualify before another skill can support Baseline.
• Opt in once to allow qualified Champions to join automatically after their current activation checks pass.
• Saved permissions survive restarts. Changed skill combinations require fresh proof, and unhealthy support falls back safely to Baseline.
• Clearer labels distinguish training progress, saved Champions, battle evidence and actual influence.

🧠 AI Coach contributions

• “Allow when ready” saves permission for proved research ideas to enter normal Challenger battles.
• A compact explanation shows the full path from research to possible Champion support.
• Coach contribution and Champion activation retain separate permissions and proof requirements.
• Improved handling of paused learning, failed refreshes and uncertain permission saves.

📈 Arena, Replay and Results

• Visible season number, duration, starting bankroll and season peak.
• Hover, touch and keyboard inspection of saved equity and cash values.
• Clearer Journey/Timeline views, chart scales and compact help.
• Per-season strategy history records which Champion skills actually influenced decisions.
• Updated README with fresh, versioned desktop and mobile screenshots.

⚡ Performance and reliability

• Urgent position, order and learning-checkpoint events receive more responsive processing during market bursts.
• Reduced training-copy memory, repeated calculations and stream-decoding overhead.
• Better pacing of optional AI and storage work.
• Fixed concurrent enrichment/pruning edge cases and improved diagnostic accuracy.
• Expanded regression coverage for activation, restarts, permissions, evidence preservation and market pressure.

Upgrading

Use Settings → Maintenance & updates → Prepare for upgrade and follow the README’s update instructions.

This release introduces database schema 15 while preserving existing data. Back up before upgrading: rollback to schema-14 releases requires restoring the matching pre-upgrade data and image together.

Signal Arcade remains paper-only. Baseline safety limits and existing coverage, uncertainty and harm checks remain enforced. Qualification and Champion status do not guarantee profit.

Documentation and upgrade instructions

v1.10.5

Signal Arcade v1.10.5 brings Champion Arena, clearer learning evidence, improved route recovery and better tools for reviewing long-running paper sessions.

🏆 Champion Arena

• Watch live comparisons and replay recorded battle checkpoints.
• Follow measured advantage, uncertainty and evidence progress.
• Stable fighter identities, varied visual styles, adaptive 3D graphics and a complete 2D fallback.
• Clear distinctions between first-Champion coronations, promotions, crown retentions and inconclusive results.
• Improved desktop/mobile layouts, graphics preferences and delayed-update handling.

Battle values come from recorded evidence. Fighter moves illustrate that evidence; they do not reconstruct market events or predict a winner.

🧠 Learning and reliability

• Clearer, separate Linear and XGBoost Entry proof, tied to each model version.
• Better background reserve validation for stale learning checkpoints and held positions.
• Fixed PumpSwap positions incorrectly blocked by an outdated quote-mint marker.
• Added verified Mayhem curve refresh and corrected fee handling.
• Improved season handover, bounded cleanup and recovery after interruption.
• Reduced unnecessary training work and made training preparation and AI Coach research more cooperative with market processing.

Missing outcomes remain unknown. The 70% coverage requirement, qualification checks and separate permission to influence trading remain intact.

📊 Diagnostics and polish

• Bounded local diagnostics history with a separate 512 MiB allowance.
• Recorded processing, training, cleanup and learning summaries for longer-term reviews.
• Clearer system status and recovery reporting.
• Updated README with fresh live screenshots in separate version folders.
• Stronger Git and Docker exclusions for local configuration, databases and generated files.

✅ Validation

• 745 backend tests and 321 frontend tests passing.
• Additional focused edge-case verification.
• Live accounting, settings preservation and route recovery checked.
• Desktop and mobile rendering verified.

⬆️ Upgrading

Keep a consistent backup and follow the README upgrade instructions. Use Settings → Maintenance & updates → Prepare for upgrade, wait for Ready, then update using the existing data volume.

Known watchpoints

Transient processing and dashboard delays have been observed and remain under review during extended operation. Monitor system status and diagnostics when running unattended.

Signal Arcade remains paper-only. Successful training or earning a Champion does not establish profitability.

v1.10.3

Signal Arcade v1.10.3 — Stable Under Load

This release improves performance and learning stability during busy Solana market streams. Signal Arcade now protects important trading and learning work more effectively while keeping the interface responsive.

Highlights

• Prioritizes held positions, pending orders, AI outcomes, and due learning checkpoints during market bursts.
• Retires stale, low-priority candidate work without silently treating incomplete evidence as valid.
• Improves queue recovery and chronological event handling under sustained public-RPC traffic.
• Moves learning recomputation away from latency-sensitive market processing.
• Safely retires completed token state only after its trading, learning, and AI work is finished.
• Replaces expensive recurring storage scans with lightweight capacity checks.
• Runs database cleanup in small, bounded chunks that yield to live market activity.
• Prevents storage maintenance from starving during continuously busy streams.
• Improves handling of transient SQLite contention and WAL checkpoint races.
• Adds clearer, compact storage-maintenance status in the interface.

Learning integrity

Baseline, linear challengers, nonlinear Entry, skill champions, and AI Coach remain aligned with the same fee-inclusive and chronological evidence rules. Interrupted or incomplete evidence cannot silently qualify a challenger.

Upgrade notes

Existing seasons, paper balances, outcomes, learning evidence, challengers, and champions remain preserved. A new season is not required after upgrading.

As always, Signal Arcade remains paper-only. Performance and champion promotion depend on the quality and quantity of observed market evidence; profit is never guaranteed.

v1.10.2

Signal Arcade v1.10.2 — Causal execution reliability

This reliability release strengthens paper-fill timing and reserve integrity without changing the Baseline, Challenger qualification rules or learning cohorts.

What’s improved

• Rebuilds live market state in authoritative Solana slot order after restarts.
• Prevents delayed lower-slot events from replacing newer reserve observations.
• Coordinates held-position RPC snapshots with the live event queue.
• Keeps reserve and event timestamps monotonic during host-clock corrections.
• Records exact RPC observation provenance for Pump and PumpSwap reserves.
• Defers any fill that would precede its reserve observation without changing the pending order, cash, positions or ledger.
• Preserves strict database validation as an independent final safeguard.
• Handles partial or invalid RPC responses without advancing causal boundaries.

Safe upgrade

Existing bankrolls, open positions, pending orders, seasons, settings, learning evidence, Challenger progress, XGBoost artifacts and Champion history are preserved.

No new season, database reset or learning reset is required when upgrading from v1.10.1.

Verification

• 386 backend tests passed
• 118 frontend tests passed
• Formatting, lint and static typing passed
• Production frontend and Docker builds passed

Signal Arcade remains a paper-only research and learning environment. It does not execute real trades or guarantee profits.

v1.10.1

Signal Arcade v1.10.1

This release strengthens paper-execution integrity, long-running reliability, and the presentation of Challenger learning.

Highlights

• Preserves per-token causal ordering even when held positions and learning outcomes receive priority during market bursts.
• Rejects lower-slot, reordered, clock-regressed, and obsolete bonding-curve events before they can affect execution or learning.
• Records the exact reserve observation, event, signature, slot, venue, quote mint, fees, and reserve values used for every new paper fill.
• Adds strict order, latency, position, venue, and receipt consistency checks.
• Detects impossible legacy execution timing without rewriting history. Affected results are preserved for audit but cannot trade, rank, compare, or qualify learning.
• Starts the clean challenger-features-v5 and learner-v6 evidence cohort.
• Keeps public-feed memory bounded while protecting unresolved Challenger and Local AI evidence.
• Improves the Learning Lab with clearer XGBoost eligibility, per-skill reigning Champions, and durable paginated battle history.
• Shows Champion-versus-contender evidence, shared outcomes, coverage, conservative results, and model families without cluttering the default view.

Safe upgrade

Upgrading from v1.9.2 or v1.10.0 preserves bankrolls, open positions, settings, seasons, evidence, and Champion history.

Older learning artifacts remain available for audit but cannot silently qualify under the new causal-evidence policy. If impossible timing is found in retained current-season receipts, Signal Arcade safely pauses that season and requests a clean new one.

Docker

docker pull nicxx2/signal-arcade:1.10.1

v1.10.0

Signal Arcade v1.10.0

Signal Arcade’s learning system is now more trustworthy, understandable and capable—while keeping the proven Baseline and all existing safety boundaries intact.

Highlights

• Added a CPU-friendly nonlinear XGBoost Entry contender alongside the Linear learner.
• XGBoost trains only after sufficient evidence and must materially outperform Linear before progressing.
• Introduced fair, restart-safe Champion tournaments using fresh common-forward outcomes.
• Added stable contender codenames, model-family labels and a durable Champion journey covering qualifications, promotions, defences and inconclusive battles.
• Improved the Challenger view with Entry, Manipulation, Sizing and Exit skill progress, current contenders, saved Champions and the first unmet proof gate.
• Clearly separated Discovery observations, authoritative Policy proof and actual Paper execution evidence.
• Strengthened exact-cohort qualification, outcome coverage, veto-quality and conservative-value requirements.
• Added bounded market-pipeline and SQLite WAL-pressure telemetry.
• Moved statistical fitting into a coalesced, single-threaded quiet-time worker for better performance on modest systems.
• Protected held positions, pending orders, AI outcomes and due learning checkpoints during heavy market bursts.
• Made learning artifacts immutable, size-bounded and digest-verified; missing or corrupted models fail closed.
• Improved first-bankroll creation feedback so busy systems remain in a clear Creating state until the authoritative result is known.
• Refreshed the Learning Lab, mobile presentation, documentation and README screenshots.

Safety and compatibility

• Existing seasons, balances, positions, outcomes and learning history remain preserved.
• Older evidence stays available for auditing but cannot silently qualify newer learning policies.
• Provider-unknown dormant inventory never receives fabricated profit or loss.
• The Baseline remains authoritative unless a Challenger passes every required proof gate and the user explicitly enables it.
• Upgrading from v1.9.2 is supported without resetting existing data.

Docker

docker pull nicxx2/signal-arcade:1.10.0

v1.9.2

Signal Arcade v1.9.2 — Stronger market integrity

This release makes Signal Arcade more selective around artificial or manipulated market activity while preserving its transparent, safety-first paper-trading philosophy.

What’s improved

• Baseline v1.5 evaluates meaningful economic participation instead of trusting raw trade counts.
• Better detection of dust-trade swarms, weak genuine volume, limited wallet participation, artificial price paths, rapid reversals and extreme trade density.
• Corroborated decisions prevent one unusual metric from blocking an otherwise promising token.
• Position sizing now requires stronger participation evidence before allocating more paper capital.
• Holding and exit logic no longer treats artificial activity as reliable support.
• Venue-local measurements prevent pre-migration and post-migration activity from being mixed incorrectly.
• Reconnects, event shedding and incomplete market windows are excluded from learning eligibility.
• Bounded rolling calculations preserve responsiveness during busy market periods.

Smarter learning

• Challenger feature schema v3 receives the expanded entry, manipulation, sizing and exit evidence.
• AI Coach and AI Decision Critic are aligned with the same integrity model while remaining shadow-only.
• Older learning cohorts and historical seasons remain preserved and cannot silently qualify using evidence they never observed.
• Champion qualification remains based on clean, fee-inclusive forward outcomes and safety gates.

Upgrade note

Existing seasons continue using the exact policy with which they started. After upgrading, begin a new season to activate Baseline v1.5 and collect fresh Challenger v3 evidence.

This remains a paper-trading research lab—not financial advice, a guaranteed scam detector or a promise of profit.

v1.9.1

🧠 Signal Arcade v1.9.1

Signal Arcade’s most capable paper-learning release yet: a stronger manipulation-aware Baseline, a skill-based statistical Challenger, a bounded AI Coach research workflow, adaptive paper sizing, and clearer season comparisons.

✨ Highlights

• Baseline v1.4 now waits for meaningful integrity evidence before entering, treats extreme isolated warnings cautiously, and uses corroborated wallet, flow, trade-structure and price-path evidence.
• Adaptive paper sizing can allocate more to mature, cleaner opportunities or reduce uncertain exploratory positions—always inside cash, exposure, reservation and price-impact limits.
• Four Challenger skills—Entry, Manipulation, Sizing and Exit—now learn, qualify and compete independently.
• Saved Champions cannot be replaced by a weaker contender. Candidates must win on common forward, fee-inclusive evidence.
• Champion Journey makes promotions, defences and inconclusive battles visible and restart-safe.
• AI Coach research lanes now study Entry, Manipulation, Sizing and Exit separately while remaining completely off the live paper path.
• A supported Coach idea still requires explicit permission, an existing matching Challenger Champion and a fresh tournament before it could contribute.
• Improved season controls provide one responsive editor for currency, bankroll, personality, drawdown policy and safe or bounded immediate transitions.
• More honest Results comparisons group seasons by currency, starting bankroll, exact profile and terminal accounting policy.
• Better dormant accounting distinguishes completed inventory, confirmed write-offs, provider uncertainty, manual endings and legacy history without inventing fills.
• Helius Economy mode reserves keyed HTTP for paced safety lookups while retaining the configured public/default WebSocket for the high-volume stream.
• Improved public-RPC burst handling catches up faster while protecting held-position, pending-order and saved-learning events.
• Live auto-season status now shows verified time remaining, paused progress, rollover, settlement and data-wait states.
• Refined desktop and mobile layouts, including clearer position evidence, responsive season selection and updated Learning views.

🛡️ Safety and learning integrity

• Signal Arcade remains paper-only—no wallet keys, signing or transaction broadcasting.
• Existing seasons, positions, learning evidence and Champions are preserved.
• Locked older seasons retain their original Baseline and policy behavior.
• Missing or incomplete evidence cannot become fabricated profit or qualification proof.
• Permanent route, mint, stale-data, stop, exposure and executable-exit protections remain active.
• Degraded learning skills are suspended without weakening the deterministic Baseline.

🐳 Docker

Multi-architecture images are available for linux/amd64 and linux/arm64:

• nicxx2/signal-arcade:1.9.1
• nicxx2/signal-arcade:latest

For an existing installation, use Settings → Maintenance & updates → Prepare for upgrade, then run:

docker compose pull

docker compose up -d

Named data and model volumes preserve paper history, learning and local AI models across the update.

See the full changelog and the refreshed README.

Signal Arcade is an educational paper-trading laboratory. Simulated results do not guarantee live profitability.

v1.8.1

Signal Arcade v1.8.1

A focused reliability patch for season continuity and honest performance comparisons.

What changed

• Automatic rollover from a legacy season now locks the new season to the current personality’s canonical profile.
• Historical legacy seasons remain correctly labelled Legacy / Unknown and are never rewritten.
• Results now compare seasons only within the same currency and exact risk profile.
• SOL, USDC, drawdown variants and different personalities remain separate in charts and performance claims.
• Mixed and legacy history stays visible without producing misleading “best season” or improvement conclusions.
• Added API metadata, compatibility handling and regression coverage for these cases.
• Updated documentation and release guidance.

Data and learning safety

• Existing seasons, trades, learning observations, Challenger models and AI Coach evidence are preserved.
• Learning remains separated by risk personality and decision-relevant configuration.
• No database reset or destructive migration is required.

Docker

Published for:

• Linux AMD64
• Linux ARM64

Image:

nicxx2/signal-arcade:1.8.1

Before upgrading, use Settings → Maintenance & updates → Prepare for upgrade, then run:

docker compose pull
docker compose up -d